Int. J. Cancer: 51,898-902 (1992) 0 1992 Wiley-Liss, Inc.
%:c
-
I.
PLOIcar on 01 rhe International Union Againsr Cancer
Puolicatton oe I’Union Internationale Contre le Cancer
ALCOHOLISM AND LIVER CIRRHOSIS IN THE ETIOLOGY OF PRIMARY LIVER CANCER Hans-Olov ADA MI'.^, Ann W. HSING~, Joseph K. MCLAUGHLIN~, Dimitrios TRICHOPOULOS3,David HACKER4,Anders EKBOM~ and Ingemar P E R S S O N ~ ~ ~ ICancer Epidemiology Unit, University Hospital, S- 751 85 Uppsala, Sweden; *BiostatisticsBranch, National Cancer Institute, Bethesda, MD; 3Departmentof Epidemioloa, Harvard School of Public Health, Boston, MA; 41nformationManagement Services, Silver Spring, MD, USA; 5Departmentof Obstetrics and Gynaecology, University Hospital, Uppsala, Sweden. The aim of this study was to determine the risk of developing primary liver cancer in patients with a diagnosis of alcoholism, liver cirrhosis, or both. Three population-based, mutually exclusive cohorts were defined on the basis of hospital discharge diagnoses between I965 and 1983. Complete follow-up through I984-excluding the first year of follow-up-showed that among 8,5 I7 patientswith a diagnosis of alcoholism, I3 cancers occutred, vs. 4.2 expected (standardized incidence ratio (SIR) = 3.1; 95% confidence interval (CI) = 1.6 to 5.3); among 3,589 patients with liver cirrhosis, 59 cancers occurred, vs. I .7 expected (SIR = 35. I; 95% CI = 26.7 to 45.3), and among 836 patients with both diagnoses, I I cancers occurred, vs. 0.3 expected (SIR = 34.3; 95% CI = 17. I to 6 I.3). Thus, alcoholism alone entailed a moderately increasedrisk and alcoholism with liver cirrhosisdid not increase the high relative risk for liver cancer more than cirrhosis alone. We conclude that alcohol intake may be a liver carcinogen only by being causally involved in the development of cirrhosis; and further, that the risk of developing liver cancer following cirrhosis in this population is similar to or higher than that after chronic hepatitis-B-virus infection in other Western countries. 0 1992 Wiley-Liss,Inc.
Alcohol intake is considered causally related to the occurrence of liver cancer (IARC, 1988), but the epidemiologic evidence is far from unequivocal (IARC, 1988; Tomatis, 1990; Austin, 1991; Hardell et al., 1984; Trichopoulos et al., 1987) and it remains uncertain whether liver cirrhosis is a necessary intermediate step in the causal pathway towards malignancy (Austin, 1991; Tuyns, 1982; Falk, 1982). Non-differential rnisclassification, notably under-reporting, of exposure is a .well-founded concern in studies of alcohol, and differential rnisclassification between cases and controls can easily arise when alcohol intake is ascertained retrospectively (Babor et al., 1987). Both types of misclassification are likely to attenuate itrue associations in most such instances. A prospective study (design should, therefore, be preferred to avoid at least the impact of differential misclassification. Our aim was to study alcoholism and liver cirrhosis as determinants of primary liver cancer risk in Sweden, a low-risk country for primary liver cancer (Whelan et al., 1990). We were especially interested in obtaining risk estimates in patients with liver cirrhosis and to assess whether undiagnosed cirrhosis might completely explain a moderately increased risk in patients with alcoholism. A population-based setting and record-linkage procedures enabled complete long-term follow-up of large cohorts. PATIENTS AND METHODS
Study population The 6-county Uppsala Health Care Region is located in central Sweden and, during the period of the study, had a total population of 1.2 to 1.3 million people. Because there is almost no private in-patient treatment in Sweden, hospital-provided medical services are in effect population-based and referable to the county in which the patient lives. From 1965 through 1983, the Swedish National Board of Health and Welfare received annual reports from all in-patient medical institutions in Sweden and recorded data on individual hospital discharges
in the In-patient Register. Besides national registration number (a unique personal identifier assigned to all Swedish citizens), each record contains data on, e.g., place of residence, hospital department, surgical procedures and up to 8 discharge diagnoses. These diagnoses were coded according to the seventh revision of the International Classification of Diseases through 1968 and according to the eighth revision thereafter. A recent publication estimated that the overall extent of under-reporting to the In-patient Register was less than 2% (Naessen et al., 1989).
The cohorts All records in the In-patient Register containing a diagnostic code for alcoholism (ICD-7:307, 322; ICD-8:291, 303) and/or liver cirrhosis (ICD-7581; ICD-8571) were considered for inclusion in the study. The national registration number allowed us to select the first recorded discharge with any of these diagnoses for each individual. A total of 14,384 patients (2,934 females and 11,450 males) were given a discharge diagnosis of alcoholism and/or cirrhosis of the liver at least once between 1965 and 1983 and were thus potentially eligible. We excluded 1,221 (8.5%) of these individuals because their entries in the Register had an incomplete or inconsistent national registration number, and thus they were not available for follow-up. Another 221 (1.5%) patients were excluded due to invalid codes or to inconsistencies revealed during the record-linkages. Thus, a total of 12,942 patients were entered into the study. Based on available discharge diagnoses, each of the patients included was allocated to 1 of the 3 mutually exclusive cohorts defined as follows: (1) patients assigned a diagnosis of alcoholism at entry but never given a diagnosis of liver cirrhosis (n = 8,517); (2) patients assigned a diagnosis of liver cirrhosis but never given a separate diagnosis of alcoholism (n = 3,589). Based on the 5-digit code of the ICD-7 and ICD-8 classification, this group could be further divided, to study whether the etiology of liver cirrhosis (alcoholism or other) influences its association with liver cancer. A total of 1,206 (34%) were coded as having cirrhosis in combination with alcoholism (“cum alcoholismo”; ICD-7581.01, 581.10; 581.20,581.99, or ICD-8:571.00,571.01,571.02,571.09,571.10, 571.11, 571.19, 571.40, 571.41); 1,082 (30%) were coded with liver cirrhosis alone (“alcoholismo non indicato”; ICD7: 581.00, ICD-8571.90, 571.98); whereas for 1,301 (36%) patients no statement was given about alcoholism (ICD-8: 571.99); (3) patients assigned the diagnosis of both alcoholism and liver cirrhosis at the same or at different discharges (n = 836). Further information on the 3 cohorts is presented in Table I. Follow-~p Record-linkage, based on the national registration number, to the nationwide Registry of Causes of Death lead to hTo whom correspondence and reprint requests should be addressed. Fax: 46-18-503431. Received: January 27,1992 and in revised form March 17,1992
899
ALCOHOLISM AND LIVER CIRRHOSIS
TABLE I - CHARACTERISTICS OF PATIENTS ASSIGNED A HOSPITAL DISCHARGE DIAGNOSIS OF ALCOHOLISM OR LIVER CIRRHOSIS IN THE UPPSALA HEALTH CARE REGION DURING 1965-1983, BY SEX Alcoholism cohort’
Number of patients Number of subsequent primary liver cancers4 Total person-years Average years of follow-up Average age at entry Average age at cancer diagnosis Average calendar year of cancer diagnosis
Liver cirrhosis cohort2
Alcoholism and liver cirrhosis cohort3
Men
Women
Men
Women
Men
Women
7606
911
1961
1628
734
102
59,289 7.8 49.5 67.5
3 6,469 7.1 49.2 59.0
91 7,648 3.9 61.6 69.6
48 7,090 4.4 64.5 70.4
16 4,156 5.7 53.4 69.1
1980
1983
1977
1977
1980
11
1
577 5.7 50.9 70.7 1984
‘Subjects had a hospital discharge diagnostic code of alcoholism (ICD-7:307, 322 or ICD-8:291, 303), but never had a diagnostic code of liver cirrhosis.-%bjects had a hospital discharge diagnostic code of liver cirrhosis (ICD-7:581 or ICD-8:571), but never had a diagnostic code of alcoholism.3Subjects with hospital discharge diagnostic codes of both alcoholism and liver cirrho~is.-~Ofthese cancers, 87 were diagnosed during the first year of follow-up. These cancers have been excluded from most of the analyses. TABLE I1 - STANDARDIZED INCIDENCE RATIOS (SIR)I FOR PRIMARY LIVER CANCERZAND BILIARY TRACT3 CANCERS AMONG PATIENTS WITH ALCOHOLISM OR LIVER CIRRHOSIS Diagnostic ETOUD~
Alcoholism cohort Liver cirrhosis cohort Alcoholism and liver cirrhosis cohort
Cancer site
Observed
Expected
SIR
95% CI
Liver Biliary tract Liver Biliary tract Liver Biliary tract
13 1 59 0 11 1
4.2 4.1 1.7 2.7 0.3 0.3
3.1 0.2 35.1 0.0 34.3 3.0
1.6-5.3 0.0-1.4 26.7-45.3 0.0-1.3 17.1-61.3 0.0-16.9
ICD-8:155.1, 155.2, 155.3, ‘Excluding the 1st year of follo~-up.-~ICD-7:155.0.-~ICD-7:155.1; 155.8,or 155.9.-4See footnotes in Table I. information on the date of death among those deceased through 1984. The National Swedish Cancer Registry, founded in 1958, was used to ascertain all incident cancers diagnosed in the cohort from start of follow-up until the end of 1984 (Swedish Cancer Registry, 1969-1987). The Cancer Registry has coded malignant diseases according to the ICD-7 classification during the entire period of study. Through 1969, only 2 4-digit codes were used, namely 155.0 (liver, primary) and 155.1 (biliary passages). From 1970, the latter code was sub-divided into 155.1 (gall bladder), 155.2 (extra-hepatic bile ducts), 155.3 (ampulla of Vater), 155.8 (multiple parts) and 155.9 (unspecified bile passages). The time of observation was calculated from the date of entry into the cohort (registration date of the first diagnosis of alcoholism or cirrhosis of the liver) until the occurrence of cancer diagnosis, death, or the end of the observation period (December 31,1984).
Statistical analysis The expected number of cancers was calculated by multiplying the number of person-years for each gender by age-specific cancer incidence rates for each 5-year age-group and calendar year of observation. These expected rates were derived from the study population, i.e., the Uppsala health care region, and included the cases observed and person-years accumulated in the cohort. For the main analyses, we excluded all person time and the 87 cases of primary liver cancer during the first year of follow-up, in order to eliminate or reduce the possible impact of selection bias. Such bias occurs in patients whose first cancer symptoms, rather than alcoholism or liver cirrhosis, gave rise to hospitalization. The standardized incidence ratio (SIR) was defined as the ratio of observed numbers of cancers to those expected. The 95 percent confidence interval (CI) of the standardized incidence
ratio was then calculated on the assumption that the observed number follows a Poisson distribution. The standard life-table method was used to calculate the cumulative incidence of primary liver cancer.
RESULTS
Because the primary aim of the study was to determine the associations between alcoholism and cirrhosis on one hand and primary liver cancer on the other, the 3-digit ICD-7 diagnostic group 155, which encompasses “biliary passages and liver, primary”, was initially separated into primary liver cancer (ICD 155.0) and biliary tract cancer (ICD7 = 155.1, ICD8 = 155.1, 155.2, 155.3, 155.8, 155.9). As shown in Table I1 the numbers of observed and expected cases of biliary tract cancers were small, and there was no evidence of an increased risk in any sub-group defined by underlying diagnosis or gender. In contrast, excess risk was consistently found for primary liver cancer. All further analyses were therefore confined to this latter group. During 1 to 18 completed years of follow-up, a total of 83 primary liver cancers were diagnosed among patients entered into the study. The diagnosis was confirmed by histopathology in 79 (95%) cases and by cytology in 3 (4%) cases, whereas in one patient clinical examination alone was reported as the diagnostic ground. Of the 83 liver cancer cases, 13 had a diagnosis of another primary cancer. Of the nine males, among the 13,2 had lung cancer, 4 prostate cancer, 1 kidney cancer, 1 skin cancer, and 1 cancer of the eye; of the four females, 1had cancer of the uterine cervix, 1 endocrine cancer, 1 multiple myeloma, and 1 skin cancer. Table I11 presents the results for the 3 cohorts by years since first admission. Among patients given a discharge diagnosis of
900
ADAMI ET AL. TABLE 111-STANDARDIZED INCIDENCE RATIOS (SIRS) FOR PRIMARY LIVER CANCER AMONG PATIENTS WITH ALCOHOLISM OR LIVER CIRRHOSIS, BY SEX AND NUMBER OF YEARS SINCE FIRST HOSPITAL ADMISSION
Completed years since -
first arlmtccmn
Men Observed
Exoected
Alcoholism cohort’ 1-4 4 5-9 6 0 10-18 1-18 10 Liver cirrhosis cohort’ 1-4 16 5-9 18 10-18 3 1-18 37 Alcoholism and liver cirrhosis cohort’ 1-4 1 5-9 8 10-18 1 1-18 10
Women SIR
95% CI
1.4 1.3 1.2 3.9
2.9 4.6 0.0 2.6
0.8-7.4 1.7-9.9 0.0-3.0 1.2-4.7
0.4 0.4 0.3 1.0
45.0 51.9 10.4 37.4
0.1 0.1 0.1 0.3
10.1 72.3 12.2 34.3
Observed
Exoected
95% C‘I
SIR
1 1 1 3
0.1 0.1 0.1 0.3
8.6 9.8 14.5 10.4
0.1-47.9 0.1-54.4 0.2-80.5 2.1-30.5
25.7-73.0 30.8-82.1 2.1-30.5 26.3-5 1.5
4 9 9 22
0.3 0.2 0.2 0.7
15.3 38.2 46.9 31.9
4.1-39.1 17.4-72.6 21.4-89.1 20.M8.3
0.1-56.2 31.1-142.5 0.2-68.0 16.4-63.1
0 0 1 1
0.0 0.0 0.01 0.03
-
-
102.2 33.8
1.3-568.6 0.4-188.3
‘See footnotes, Table I. alcoholism at entry to the cohort without any mention of cirrhosis at the same or any preceding or subsequent hospitalization, a total of 13 primary liver cancers occurred in men and women vs. 4.2 expected (SIR = 3.1; 95% CL1.6 to 5.3). No clear temporal trend was seen with years of follow-up, although relative risk estimates by gender and years of follow-up were unstable due to small numbers. Patients with liver cirrhosis showed markedly elevated risk (Table 111). Eighty of the 87 liver cancers found in the first year of follow-up were from this cohort, most probably due to :selectionbias. One year and later after start of follow-up, 59 cancers occurred in men and women, vs. 1.7 expected for an SIR of 35.1 (95% CI:26.7 to 45.3). Overall relative risk was similar in both sexes (SIR = 37.4 for men; SIR = 31.9 for women), but there was some evidence of higher relative risk after longer follow-up in women with liver cirrhosis. Among patients with both diagnoses of alcoholism and cirrhosis at the same or at subsequent hospital discharges, the SIR, based on 17 cases (16 men and one women) was 34.3 (95% CI: 17.1-61.3) and was similar to that observed in the cirrhosis cohort (Table 111). Figure 1 shows the cumulative incidence of liver cancer in all patients with liver cirrhosis (with or without alcoholism). When the first year of follow-up was excluded, the cumulative incidence (with 95% confidence interval) was 1.1 (0.6 to 1.6) “0, 4.2 (3.0 to 5.3) %, and 5.9 (4.2 to 7.5) % after 5, 10 and 15 years respectively. The cumulative rates were about 2% higher when no latency period was applied. Age at first diagnosis of alcoholism was seemingly not an important determinant of the relative risk of developing primary liver cancer, with no clear trend by age in either sex (Table IV). In patients with liver cirrhosis, the relative risk decreased with increasing age at first diagnosis, both in men and in women (Table IV). The cohort of patients with liver cirrhosis but without a separate diagnostic code for alcoholism was further subdivided as described in “Patients and Methods” and further specified in Table V. The relative risks of developing primary liver cancer were similar in patients coded as having liver cirrhosis with and without alcoholism. DISCUSSION
This prospective study indicates that alcoholism without clinically evident cirrhosis entails only a moderately increased
“1 12
O
~
0
., .
.
.
,
5
.
.
.
.
,
.
10 Years of Follow-up
.
.
.
,
15
,
.
.
,
FIGURE 1 - Cumulative incidence of primary liver cancer (ICD-7 code 155.0) among 4,425 patients with liver cirrhosis. A total of 156 primary liver cancers occurred; 70 of them were included when a one-year latency was applied. Vertical bars indicate 95 percent confidence intervals.
risk of developing primary liver cancer. In contrast, patients with liver cirrhosis experienced an excess risk more than 10 times higher than those with alcoholism alone. The strength of this study is the population-based design, the completeness of follow-up, and the size, which allows fairly stable risk estimation. In a separate analysis, support for this study design was provided when cancer sites other than liver were examined in the same study groups, and the well-established associations of alcohol with oral, esophageal, and laryngeal cancers were observed with 3- to 7-fold excess risks in both sexes (data not shown). However, there are also limitations in these data, namely some non-differential misclassification of exposure (alcoholism and liver cirrhosis) and lack of information on potential confounding and interacting factors. With regard to misclassification, alcohol abuse accounts for a major proportion of all cases of liver cirrhosis occurring in Western populations (Tuyns, 1982). Thus, many of the patients with cirrhosis but without any discharge diagnosis revealing alcohol abuse might in fact be high consumers, or even alcoholics, as further indicated by detailed analysis of ICD codes for liver cirrhosis (Table V). Likewise, in the alcoholism cohort, the prevalence of cirrhosis is probably higher than in the background population; the disease might still be pre-
901
ALCOHOLISM AND LIVER CIRRHOSIS
TABLE N - STANDARDIZED INCIDENCE RATIOS (SIRS)' FOR PRIMARY LIVER CANCER AMONG PATIENTS WITH ALCOHOLISM AND LIVER CIRRHOSIS BY SEX AND AGE AT FIRST HOSPITAL ADMISSION FOR ALCOHOLISM OR LIVER CIRRHOSIS (FIRST YEAR OF FOLLOW-UP EXCLUDED) Age at first hospital admission for alcoholism or liver cirrhosis
Men N
~
~
~ Observed , f
Alcoholism cohort 1 3872 < 50 1785 2 50-59 5 1298 60-69 65 1 2 70 + Liver cirrhosis cohort < 50 384 3 50-59 445 9 60-69 528 18 604 7 70 + Alcoholism and liver cirrhosis cohort 285 1 < 50 221 1 50-59 60-69 171 7 57 1 70 +
Expected
Women SIR
95% CI
N:zktPfObserved
Expected
SIR
95% CI
0.5 1.2 1.4 0.8
2.0 1.7 3.5 2.4
0.0-11.1 0.2-6.2 1.1-8.1 0.3-8.8
467 219 133 92
2 0 1 0
0.05 0.1 0.1 0.1
42.8 0.0 11.8 0.0
4.8-154.5 0.0-56.2 0.2-65.6 0.040.4
0.04 0.2 0.4 0.4
82.8 50.9 48.1 17.4
16.6-241.8 23.3-96.8 28.5-76.1 7.0-35.8
278 26 1 402 687
5
4 10 3
0.03 0.1 0.2 0.4
181.7 50.4 44.4 8.4
58.6-424.0 13.6-128.9 21.2-81.6 1.7-24.6
0.0 0.1 0.1 0.03
25.0 10.7 56.0 30.3
0.3-139.1 0.1-59.5 22.4-115.5 0.4-168.6
51 29 13 9
0 1 0 0
0.0 0.01 0.01 0.01
0.0 173.7 0.0 0.0
0.0-762.1 2.3-966.3 0.0-392.9 0.0-380.3
'See footnotes. Table I. TABLE V - STANDARDIZED INCIDENCE RATIOS (SIRS)' FOR PRIMARY LIVER CANCER AMONG PATIENTS WITH LIVER CIRRHOSIS BY ALCOHOLISM STATUS. ACCORDING TO THE 5-DIGIT ICD CODE Men
Women
Observed ExDected SIR
Liver cirrhosis without alcoholism2 Liver cirrhosis with alcoholism3 Liver cirrhosis with unknown alcoholism4
95% CI
Observed Exoected SIR
95% CI
9
0.2
42.1 19.2-80.0
14
0.4
36.9 20.2-62.0
18
0.4
42.0 24.9-66.4
3
0.1
48.8 9.8-142.5
10
0.3
28.7 13.8-52.8
5
0.2
20.1 6.546.9
'Excluding cases of primary liver cancer from the first year of follow-up.-*ICD-7:581.00 or ICD-8571.90 or 571.98; 391 men and 691 ~omen.-~ICD-7:581.01,581.10, 581.20, 581.99 or ICD-8:571.00,571.02,571.09,571.10,571.11,571.19,571.40,571.41; 937 men and 269 ~omen.-~ICD8571.99; 633 men and 668 women. clinical, or it might have been diagnosed previously in another health care region (if the patient has moved into the Uppsala Health Care Region) before the start of the register in 1965 or on an out-patient basis. Under such circumstances, hepatic cirrhosis might not appear as a discharge diagnosis if a patient has been hospitalized, eg., due to an accident. Misclassification of the exposure (alcoholism or cirrhosis) entails underestimation of the true difference in risk between the diagnostic groups. Given the relative magnitude of the respective relative risks, it is evident that the existence of alcoholism per se could have influenced the risk estimates for cirrhosis only marginally. This notion is further supported by the similar excess risk in patients with both diagnoses and in various diagnostic sub-groups of liver cirrhosis (Table V). In contrast, liver cirrhosis might introduce major bias when the association between alcoholism and liver cancer is studied. Indeed, if the relative risk for alcoholism alone is 3 and for liver cirrhosis 36, then 6 percent of undetectable cirrhosis would completely explain the alcoholism association. The impact of confounding by other factors is unknown. In studies with more complete assessment of exposure, confounding seemed to be a minor problem in analyses of alcohol and liver cancer (Tomatis, 1990; La Vecchia et al., 1990), although tobacco smoking, which is more common among alcoholics (Kozlowski and Ferrence, 1990; Cederlof et al., 1975), has been found to be a moderate risk factor for liver cancer (Austin, 1991; Trichopoulos et al., 1987; Hsing et al., 1990). In the present study, the relative risk for lung cancer (excluding the first year and combining sexes) was 2.1 and 1.1among patients with alcoholism and liver cirrhosis, respectively (data not
shown). Hence, tobacco smoking cannot explain the strong associations observed in our study. For liver cirrhosis, the prime candidate as a confounder or an effect modifier of the association between cirrhosis and liver cancer would be chronic active hepatitis-B (Tomatis, 1990; Trichopoulos et aL, 1987) and perhaps hepatitis-C (Kaklamani et al., 1991) infection. The prevalence of the former condition is, however, low in the Swedish population (Olsson et al., 1988), whereas that of the latter is largely unknown. The prevalence of both might be higher in patients with alcoholism and notably liver cirrhosis, on account of these patients' special living conditions, lower socio-economic status and more frequent contact with health care. It is unlikely, though, that the impact of such factors could explain more than a small part of the difference in excess relative risk for primary liver cancer between alcoholism patients with and without liver cirrhosis. In fact, epidemiologic studies in Western populations have in general shown more modest, 10- to 20-fold increases of the risk of primary liver cancer following chronic hepatitis-B infection (Trichopoulos et al., 1987; La Vecchia et al., 1990; Kaklamani et al., 1991; Yu et al., 1983) than we found in patients with liver cirrhosis. This reasoning leads us to the tentative conclusion that alcohol and even alcoholism leads to liver cancer mainly (perhaps exclusively) through development of liver cirrhosis as an intermediate step. With regard to liver cirrhosis and liver cancer, methodologic concerns may be less serious and causal interpretation more straightforward. Since ascertainment of both hepatic and biliary-tract cancer might be more complete in patients with cirrhosis, thereby exaggerating relative risk estimates,it was reassur-
902
ADAMI ET AL.
ing to find little or no evidence of excess risk in relation to biliary cancer. In Sweden, as in Denmark (Skinhoj and Prytz, 1981), alcoholism and unknown (“idiopathic”) causes likely account for most cases of livcr cirrhosis, whereas the proportion due to chronic viral hepatitis as suggested above is low. Previous follow-up studies of primary liver canccr in paticnts with alcoholism were small, with few observed cases and generally reported risks about 50% higher than in nonalcoholics (Tomatis, 1990; Austin, 1991; Hardell et al., 1984; Trichopoulos et al., 1987; Tuyns, 1982; Falk, 1982; Babor et al., 1987). Conceivably, our higher estimates may b e due to more severe and long-lasting abuse (and consequently a higher prcvalcnce of liver cirrhosis) in those given a discharge diagnosis of alcoholism. This may pertain especially to women, for whom social barriers entail a more restrictive use of alcoholism as a diagnosis. However, even in men the diagnosis is restricted to those with obvious alcoholism and not uscd for patients who have accidentally abused. Tndeed, underdiagnosis of alcoholism in the dischargc rcgister is likely to be substantial. Risk estimates for livcr cancer among patients with liver cirrhosis provided by case-control studies are generally lower
(Trichopoulos et al., 1987; La Vecchia et al., 1990), cvcn whcn no latency period was applied to avoid sclection bias (La Vccchia et al., 1990). Prospective studies in paticnts with cirrhosis have bccn small in siLe (Zaman et al., 1985; Hadengue et al., 1990; Kobayachi et al., 1990; Oka et al., 1990; Borzio et al., 1991) and quantitative risk estimates were not always rcportcd (Kobayachi et al., 1990; Oka et al., 1990). Our data did not support previous claims that mcn (Zaman et ul., 1985; Kobayashi et al., 1990) or age over 60 (Borzio et UL, 1991) are effect modifiers for liver cancer in paticnts with cirrhosis. It has been reported that 10 to 30% of patients with cirrhosis develop liver cancer (Rothman, 1980); that the cumulative incidence after 6 years is 39% (Oka et al., 1990) and that 213 (Okudaet al., 1987) of patients with liver cirrhosis in Japan dic from liver cancer. These extreme rates clearly did not pertain to the Swedish population (Fig. 1 ) . ACKNOWLEDGEMENTS This study was supported by grants from the Swedish Cancer Socicty. W e thank Mr. E. Berger for excellent assistance with the data analyses.
REFEKENCLS ALSTIN.H., The role of tobacco use and alcohol consumption in the etiology of hepatocellular carcinoma. In: E. Tabor. A.M. Di Bisceglie and R.H. Purcell (eds.), Etiology, pathology, and treatment ofhepatocellular carcinoma in North America. Houston: Adv. appl. Biotechnol.,13, pp. 57-75 (1991). BABOR, T.. STEVENS, R. and MARLATT, A,, Verbal report methods in clinical research on alcoholism: response bias and its minimization.J. Stud. Alcohol, 48,410-424 (1987). BORZIO,M., BRUNO, S., RONCALLI, M., MELS,G.C., RAMELLA, G., BORLIO,F., LEANDRO, G. and PODDA,M., Liver cell dysplasia and risk of hepatocellular carcinoma in cirrhosis: a preliminary report. Brit. med. J., 302,1312 (1991). CEDERI.~F, R., FRIBERG, L., HRUBEC,Z. and LOKICH,U., The relationship of vmoking and some social co-variables to mortality and cancer morbidity. The Department of Environmental Hygiene, the Karolinska Institute, Stockholm (1975). FALK, H.. Liver. In: D. Schottenfeld and J.F. Fraumeni (eds.), Cancer epidemiology and prevention, pp. 668-682, Saunders, Philadelphia (1982). HADENG~JE, A., N’DRI, N. and BENHAMON, J.P., Relative risk of hepatocellular carcinoma in HBsAg positive vs. alcoholic cirrhosis. A cross-sectional study. Liver, 10, 147-151 (1990). HARDELL, I ,., BENGTSSON,N.O., JONSSON, U., ERIKSSOK. S. and LARSSON, L.G., Aetiological aspects on primary liver cancer with special regard to alcohol, organic solvents and acute intermittent porphyria- an epidemiological investigation. Brit. J. Cancer, 50, 389397 (1984). HSING,A.W.. MCLAUGHLIN, J.K., HRUBEC,Z., BLOT,W.J. and FRAUMENI, J.F., Cigarette smoking and liver cancer among US veterans. Cancer Causes Control, 1,217-221 (1990). INTERNATIONALA ~ ~ L NFOR C YRESEARCH ON CANCER,Alcohol drinking. IARC Monographs on the evaluation of carcinogenic risks, 44, TARC, Lyon (1988). KAKLAMANI. E.. TRICHOPOULOS, D., Tzon-ou, A., ZAVITSANOS, X.. KOUMANTAKI, Y., HKIZAKIS. A., HSIEH,C.C. and HXI‘ZIYANILIS, S., Hepatitis B and C viruses and their interaction in the origin of hepatocellular carcinoma.J. Anter. med. Ass., 265, 1974-1976 (1991). KOBAYASHI, K., UNOURA,M.? TAYAKA, N. and HATTORI, N., A comparison bctwccn hepatocellular-carcinoma-developing and noncarcinoma-developing patients with cirrhosis over a long follow-up 37,445-448 (1990). period. Hepatogastroenter[~lo~, KOZLOWSKI, L.T. and F~KKE~YICE, R.G., Statistical control in research on alcohol and tobacco: an example from research on alcohol and mortality. Brit. J. Addiction, 85,271-278 (1990).
LA VECCHTA, C., NEGRI,E., D’AvA~Lo,B., BOYLE, P. and FRANCESCHI,S., Medical history and primary liver cancer. Cancer Rex, 50, 6274-6277 (1990). NAESSEN, T., PARKER, R., PERSSON, 1.; ZACK,M. and ADAMI, H.O., Time trends in incidence rates of first hip fracture in the Uppsala Health Care Region, Sweden 1965-1983. Amer. J. Epidemiol., 130, 289-299 (1989). OM, €I,, KLRIOKA,N., KIM,K., K?L”O,T., KUROKJ.T., MIZOGUCHI, Y. and KOBAYASHI; K.; Prospective study of early detection of hepatocellular carcinoma in patients with cirrhosis. Hepatology, 12, 680-687 (1990). OKUDA,K.: FUIIMOTO, I., HANAI,LA. and URANO,Y.. Changing incidence of hepatocellular carcinoma in Japan. Cancer Rex, 47, 49674972 (1987). OLSSON. R., LINDBERG, J., WEILANU, O., NILSSON,L. and members of the Swedish Internal Medicine Club and Clinics for Infectious Diseases, Chronic active hepatitis in Sweden. Scund. J. Gastroenterol., 23, 463470 (1988). ROTHMAN, K.J., The proportion of cancer attributable to alcohol consumption.Prev. Med., 9,174-179 (1980). SKINHBJ, P. and PRYTZ,H., Changing mortality from cirrhosis in Denmark 1965-1978. Scund. J. Gastroenterol., 16,833-837 (1981). SWEDISH CANCER REGISTRY, Cancer incidence in Sweden I965-1984. National Board of Health and Welfare (Annual publications) Stockholm, Sweden (1969-1987). TOMATIS, L., Cancer cuuses, occurrence and control. IARC Scientific Publication 100, IARC, Lyon (1990). TRICHOPOULOS, D., DAY,N.E., KCIKLAMANI, E., TZONOU, A.! MuGoz, N., ZAVIISANOS, X., KOUMAN IAKI, Y. and TRICHOPOULOS, A,, Hepatitis B virus, tobacco smoking and ethanol consumption in the etiology of hepatocellular carcinoma. Int. J. Cancer, 39,45-49 (1987). TLYNS,A.J.. Alcohol. In: D. Schottenfeld, D. and J.F. Fraumcni (eds.), Cancer epidemiology and prevention, pp. 293-303, Saunders, Philadelphia (1982). WHELAN, S.L., PARKIN, D.M. and MASUYF,R, E. (eds.), Patterns of cancer in$ve continents. IARC ScientificPublication 102, IARC, Lyon (1990). Yu, M.C., MACK,T., IIANISCH? R., PETERS,R.L., HENDERSON, B.E. and PIKE,M.C., Hepatitis, alcohol consumption, cigarette smoking, and hepatocellular carcinoma in Los Angeles. Cancer Rex, 43, 60776079 (1983). ZAMAN, S.N., MELIA,W .M.,JOHNSON. R.D., PORTMAN, R.C., JOHKSON, P.3. and WILLIAMS, R., Risk factors in development of hepatocellular carcinoma in cirrhosis. Prospective study of 613 patients. Lancet, I, 1357-1360 (1985).
%:c
-
I.
PLOIcar on 01 rhe International Union Againsr Cancer
Puolicatton oe I’Union Internationale Contre le Cancer
ALCOHOLISM AND LIVER CIRRHOSIS IN THE ETIOLOGY OF PRIMARY LIVER CANCER Hans-Olov ADA MI'.^, Ann W. HSING~, Joseph K. MCLAUGHLIN~, Dimitrios TRICHOPOULOS3,David HACKER4,Anders EKBOM~ and Ingemar P E R S S O N ~ ~ ~ ICancer Epidemiology Unit, University Hospital, S- 751 85 Uppsala, Sweden; *BiostatisticsBranch, National Cancer Institute, Bethesda, MD; 3Departmentof Epidemioloa, Harvard School of Public Health, Boston, MA; 41nformationManagement Services, Silver Spring, MD, USA; 5Departmentof Obstetrics and Gynaecology, University Hospital, Uppsala, Sweden. The aim of this study was to determine the risk of developing primary liver cancer in patients with a diagnosis of alcoholism, liver cirrhosis, or both. Three population-based, mutually exclusive cohorts were defined on the basis of hospital discharge diagnoses between I965 and 1983. Complete follow-up through I984-excluding the first year of follow-up-showed that among 8,5 I7 patientswith a diagnosis of alcoholism, I3 cancers occutred, vs. 4.2 expected (standardized incidence ratio (SIR) = 3.1; 95% confidence interval (CI) = 1.6 to 5.3); among 3,589 patients with liver cirrhosis, 59 cancers occurred, vs. I .7 expected (SIR = 35. I; 95% CI = 26.7 to 45.3), and among 836 patients with both diagnoses, I I cancers occurred, vs. 0.3 expected (SIR = 34.3; 95% CI = 17. I to 6 I.3). Thus, alcoholism alone entailed a moderately increasedrisk and alcoholism with liver cirrhosisdid not increase the high relative risk for liver cancer more than cirrhosis alone. We conclude that alcohol intake may be a liver carcinogen only by being causally involved in the development of cirrhosis; and further, that the risk of developing liver cancer following cirrhosis in this population is similar to or higher than that after chronic hepatitis-B-virus infection in other Western countries. 0 1992 Wiley-Liss,Inc.
Alcohol intake is considered causally related to the occurrence of liver cancer (IARC, 1988), but the epidemiologic evidence is far from unequivocal (IARC, 1988; Tomatis, 1990; Austin, 1991; Hardell et al., 1984; Trichopoulos et al., 1987) and it remains uncertain whether liver cirrhosis is a necessary intermediate step in the causal pathway towards malignancy (Austin, 1991; Tuyns, 1982; Falk, 1982). Non-differential rnisclassification, notably under-reporting, of exposure is a .well-founded concern in studies of alcohol, and differential rnisclassification between cases and controls can easily arise when alcohol intake is ascertained retrospectively (Babor et al., 1987). Both types of misclassification are likely to attenuate itrue associations in most such instances. A prospective study (design should, therefore, be preferred to avoid at least the impact of differential misclassification. Our aim was to study alcoholism and liver cirrhosis as determinants of primary liver cancer risk in Sweden, a low-risk country for primary liver cancer (Whelan et al., 1990). We were especially interested in obtaining risk estimates in patients with liver cirrhosis and to assess whether undiagnosed cirrhosis might completely explain a moderately increased risk in patients with alcoholism. A population-based setting and record-linkage procedures enabled complete long-term follow-up of large cohorts. PATIENTS AND METHODS
Study population The 6-county Uppsala Health Care Region is located in central Sweden and, during the period of the study, had a total population of 1.2 to 1.3 million people. Because there is almost no private in-patient treatment in Sweden, hospital-provided medical services are in effect population-based and referable to the county in which the patient lives. From 1965 through 1983, the Swedish National Board of Health and Welfare received annual reports from all in-patient medical institutions in Sweden and recorded data on individual hospital discharges
in the In-patient Register. Besides national registration number (a unique personal identifier assigned to all Swedish citizens), each record contains data on, e.g., place of residence, hospital department, surgical procedures and up to 8 discharge diagnoses. These diagnoses were coded according to the seventh revision of the International Classification of Diseases through 1968 and according to the eighth revision thereafter. A recent publication estimated that the overall extent of under-reporting to the In-patient Register was less than 2% (Naessen et al., 1989).
The cohorts All records in the In-patient Register containing a diagnostic code for alcoholism (ICD-7:307, 322; ICD-8:291, 303) and/or liver cirrhosis (ICD-7581; ICD-8571) were considered for inclusion in the study. The national registration number allowed us to select the first recorded discharge with any of these diagnoses for each individual. A total of 14,384 patients (2,934 females and 11,450 males) were given a discharge diagnosis of alcoholism and/or cirrhosis of the liver at least once between 1965 and 1983 and were thus potentially eligible. We excluded 1,221 (8.5%) of these individuals because their entries in the Register had an incomplete or inconsistent national registration number, and thus they were not available for follow-up. Another 221 (1.5%) patients were excluded due to invalid codes or to inconsistencies revealed during the record-linkages. Thus, a total of 12,942 patients were entered into the study. Based on available discharge diagnoses, each of the patients included was allocated to 1 of the 3 mutually exclusive cohorts defined as follows: (1) patients assigned a diagnosis of alcoholism at entry but never given a diagnosis of liver cirrhosis (n = 8,517); (2) patients assigned a diagnosis of liver cirrhosis but never given a separate diagnosis of alcoholism (n = 3,589). Based on the 5-digit code of the ICD-7 and ICD-8 classification, this group could be further divided, to study whether the etiology of liver cirrhosis (alcoholism or other) influences its association with liver cancer. A total of 1,206 (34%) were coded as having cirrhosis in combination with alcoholism (“cum alcoholismo”; ICD-7581.01, 581.10; 581.20,581.99, or ICD-8:571.00,571.01,571.02,571.09,571.10, 571.11, 571.19, 571.40, 571.41); 1,082 (30%) were coded with liver cirrhosis alone (“alcoholismo non indicato”; ICD7: 581.00, ICD-8571.90, 571.98); whereas for 1,301 (36%) patients no statement was given about alcoholism (ICD-8: 571.99); (3) patients assigned the diagnosis of both alcoholism and liver cirrhosis at the same or at different discharges (n = 836). Further information on the 3 cohorts is presented in Table I. Follow-~p Record-linkage, based on the national registration number, to the nationwide Registry of Causes of Death lead to hTo whom correspondence and reprint requests should be addressed. Fax: 46-18-503431. Received: January 27,1992 and in revised form March 17,1992
899
ALCOHOLISM AND LIVER CIRRHOSIS
TABLE I - CHARACTERISTICS OF PATIENTS ASSIGNED A HOSPITAL DISCHARGE DIAGNOSIS OF ALCOHOLISM OR LIVER CIRRHOSIS IN THE UPPSALA HEALTH CARE REGION DURING 1965-1983, BY SEX Alcoholism cohort’
Number of patients Number of subsequent primary liver cancers4 Total person-years Average years of follow-up Average age at entry Average age at cancer diagnosis Average calendar year of cancer diagnosis
Liver cirrhosis cohort2
Alcoholism and liver cirrhosis cohort3
Men
Women
Men
Women
Men
Women
7606
911
1961
1628
734
102
59,289 7.8 49.5 67.5
3 6,469 7.1 49.2 59.0
91 7,648 3.9 61.6 69.6
48 7,090 4.4 64.5 70.4
16 4,156 5.7 53.4 69.1
1980
1983
1977
1977
1980
11
1
577 5.7 50.9 70.7 1984
‘Subjects had a hospital discharge diagnostic code of alcoholism (ICD-7:307, 322 or ICD-8:291, 303), but never had a diagnostic code of liver cirrhosis.-%bjects had a hospital discharge diagnostic code of liver cirrhosis (ICD-7:581 or ICD-8:571), but never had a diagnostic code of alcoholism.3Subjects with hospital discharge diagnostic codes of both alcoholism and liver cirrho~is.-~Ofthese cancers, 87 were diagnosed during the first year of follow-up. These cancers have been excluded from most of the analyses. TABLE I1 - STANDARDIZED INCIDENCE RATIOS (SIR)I FOR PRIMARY LIVER CANCERZAND BILIARY TRACT3 CANCERS AMONG PATIENTS WITH ALCOHOLISM OR LIVER CIRRHOSIS Diagnostic ETOUD~
Alcoholism cohort Liver cirrhosis cohort Alcoholism and liver cirrhosis cohort
Cancer site
Observed
Expected
SIR
95% CI
Liver Biliary tract Liver Biliary tract Liver Biliary tract
13 1 59 0 11 1
4.2 4.1 1.7 2.7 0.3 0.3
3.1 0.2 35.1 0.0 34.3 3.0
1.6-5.3 0.0-1.4 26.7-45.3 0.0-1.3 17.1-61.3 0.0-16.9
ICD-8:155.1, 155.2, 155.3, ‘Excluding the 1st year of follo~-up.-~ICD-7:155.0.-~ICD-7:155.1; 155.8,or 155.9.-4See footnotes in Table I. information on the date of death among those deceased through 1984. The National Swedish Cancer Registry, founded in 1958, was used to ascertain all incident cancers diagnosed in the cohort from start of follow-up until the end of 1984 (Swedish Cancer Registry, 1969-1987). The Cancer Registry has coded malignant diseases according to the ICD-7 classification during the entire period of study. Through 1969, only 2 4-digit codes were used, namely 155.0 (liver, primary) and 155.1 (biliary passages). From 1970, the latter code was sub-divided into 155.1 (gall bladder), 155.2 (extra-hepatic bile ducts), 155.3 (ampulla of Vater), 155.8 (multiple parts) and 155.9 (unspecified bile passages). The time of observation was calculated from the date of entry into the cohort (registration date of the first diagnosis of alcoholism or cirrhosis of the liver) until the occurrence of cancer diagnosis, death, or the end of the observation period (December 31,1984).
Statistical analysis The expected number of cancers was calculated by multiplying the number of person-years for each gender by age-specific cancer incidence rates for each 5-year age-group and calendar year of observation. These expected rates were derived from the study population, i.e., the Uppsala health care region, and included the cases observed and person-years accumulated in the cohort. For the main analyses, we excluded all person time and the 87 cases of primary liver cancer during the first year of follow-up, in order to eliminate or reduce the possible impact of selection bias. Such bias occurs in patients whose first cancer symptoms, rather than alcoholism or liver cirrhosis, gave rise to hospitalization. The standardized incidence ratio (SIR) was defined as the ratio of observed numbers of cancers to those expected. The 95 percent confidence interval (CI) of the standardized incidence
ratio was then calculated on the assumption that the observed number follows a Poisson distribution. The standard life-table method was used to calculate the cumulative incidence of primary liver cancer.
RESULTS
Because the primary aim of the study was to determine the associations between alcoholism and cirrhosis on one hand and primary liver cancer on the other, the 3-digit ICD-7 diagnostic group 155, which encompasses “biliary passages and liver, primary”, was initially separated into primary liver cancer (ICD 155.0) and biliary tract cancer (ICD7 = 155.1, ICD8 = 155.1, 155.2, 155.3, 155.8, 155.9). As shown in Table I1 the numbers of observed and expected cases of biliary tract cancers were small, and there was no evidence of an increased risk in any sub-group defined by underlying diagnosis or gender. In contrast, excess risk was consistently found for primary liver cancer. All further analyses were therefore confined to this latter group. During 1 to 18 completed years of follow-up, a total of 83 primary liver cancers were diagnosed among patients entered into the study. The diagnosis was confirmed by histopathology in 79 (95%) cases and by cytology in 3 (4%) cases, whereas in one patient clinical examination alone was reported as the diagnostic ground. Of the 83 liver cancer cases, 13 had a diagnosis of another primary cancer. Of the nine males, among the 13,2 had lung cancer, 4 prostate cancer, 1 kidney cancer, 1 skin cancer, and 1 cancer of the eye; of the four females, 1had cancer of the uterine cervix, 1 endocrine cancer, 1 multiple myeloma, and 1 skin cancer. Table I11 presents the results for the 3 cohorts by years since first admission. Among patients given a discharge diagnosis of
900
ADAMI ET AL. TABLE 111-STANDARDIZED INCIDENCE RATIOS (SIRS) FOR PRIMARY LIVER CANCER AMONG PATIENTS WITH ALCOHOLISM OR LIVER CIRRHOSIS, BY SEX AND NUMBER OF YEARS SINCE FIRST HOSPITAL ADMISSION
Completed years since -
first arlmtccmn
Men Observed
Exoected
Alcoholism cohort’ 1-4 4 5-9 6 0 10-18 1-18 10 Liver cirrhosis cohort’ 1-4 16 5-9 18 10-18 3 1-18 37 Alcoholism and liver cirrhosis cohort’ 1-4 1 5-9 8 10-18 1 1-18 10
Women SIR
95% CI
1.4 1.3 1.2 3.9
2.9 4.6 0.0 2.6
0.8-7.4 1.7-9.9 0.0-3.0 1.2-4.7
0.4 0.4 0.3 1.0
45.0 51.9 10.4 37.4
0.1 0.1 0.1 0.3
10.1 72.3 12.2 34.3
Observed
Exoected
95% C‘I
SIR
1 1 1 3
0.1 0.1 0.1 0.3
8.6 9.8 14.5 10.4
0.1-47.9 0.1-54.4 0.2-80.5 2.1-30.5
25.7-73.0 30.8-82.1 2.1-30.5 26.3-5 1.5
4 9 9 22
0.3 0.2 0.2 0.7
15.3 38.2 46.9 31.9
4.1-39.1 17.4-72.6 21.4-89.1 20.M8.3
0.1-56.2 31.1-142.5 0.2-68.0 16.4-63.1
0 0 1 1
0.0 0.0 0.01 0.03
-
-
102.2 33.8
1.3-568.6 0.4-188.3
‘See footnotes, Table I. alcoholism at entry to the cohort without any mention of cirrhosis at the same or any preceding or subsequent hospitalization, a total of 13 primary liver cancers occurred in men and women vs. 4.2 expected (SIR = 3.1; 95% CL1.6 to 5.3). No clear temporal trend was seen with years of follow-up, although relative risk estimates by gender and years of follow-up were unstable due to small numbers. Patients with liver cirrhosis showed markedly elevated risk (Table 111). Eighty of the 87 liver cancers found in the first year of follow-up were from this cohort, most probably due to :selectionbias. One year and later after start of follow-up, 59 cancers occurred in men and women, vs. 1.7 expected for an SIR of 35.1 (95% CI:26.7 to 45.3). Overall relative risk was similar in both sexes (SIR = 37.4 for men; SIR = 31.9 for women), but there was some evidence of higher relative risk after longer follow-up in women with liver cirrhosis. Among patients with both diagnoses of alcoholism and cirrhosis at the same or at subsequent hospital discharges, the SIR, based on 17 cases (16 men and one women) was 34.3 (95% CI: 17.1-61.3) and was similar to that observed in the cirrhosis cohort (Table 111). Figure 1 shows the cumulative incidence of liver cancer in all patients with liver cirrhosis (with or without alcoholism). When the first year of follow-up was excluded, the cumulative incidence (with 95% confidence interval) was 1.1 (0.6 to 1.6) “0, 4.2 (3.0 to 5.3) %, and 5.9 (4.2 to 7.5) % after 5, 10 and 15 years respectively. The cumulative rates were about 2% higher when no latency period was applied. Age at first diagnosis of alcoholism was seemingly not an important determinant of the relative risk of developing primary liver cancer, with no clear trend by age in either sex (Table IV). In patients with liver cirrhosis, the relative risk decreased with increasing age at first diagnosis, both in men and in women (Table IV). The cohort of patients with liver cirrhosis but without a separate diagnostic code for alcoholism was further subdivided as described in “Patients and Methods” and further specified in Table V. The relative risks of developing primary liver cancer were similar in patients coded as having liver cirrhosis with and without alcoholism. DISCUSSION
This prospective study indicates that alcoholism without clinically evident cirrhosis entails only a moderately increased
“1 12
O
~
0
., .
.
.
,
5
.
.
.
.
,
.
10 Years of Follow-up
.
.
.
,
15
,
.
.
,
FIGURE 1 - Cumulative incidence of primary liver cancer (ICD-7 code 155.0) among 4,425 patients with liver cirrhosis. A total of 156 primary liver cancers occurred; 70 of them were included when a one-year latency was applied. Vertical bars indicate 95 percent confidence intervals.
risk of developing primary liver cancer. In contrast, patients with liver cirrhosis experienced an excess risk more than 10 times higher than those with alcoholism alone. The strength of this study is the population-based design, the completeness of follow-up, and the size, which allows fairly stable risk estimation. In a separate analysis, support for this study design was provided when cancer sites other than liver were examined in the same study groups, and the well-established associations of alcohol with oral, esophageal, and laryngeal cancers were observed with 3- to 7-fold excess risks in both sexes (data not shown). However, there are also limitations in these data, namely some non-differential misclassification of exposure (alcoholism and liver cirrhosis) and lack of information on potential confounding and interacting factors. With regard to misclassification, alcohol abuse accounts for a major proportion of all cases of liver cirrhosis occurring in Western populations (Tuyns, 1982). Thus, many of the patients with cirrhosis but without any discharge diagnosis revealing alcohol abuse might in fact be high consumers, or even alcoholics, as further indicated by detailed analysis of ICD codes for liver cirrhosis (Table V). Likewise, in the alcoholism cohort, the prevalence of cirrhosis is probably higher than in the background population; the disease might still be pre-
901
ALCOHOLISM AND LIVER CIRRHOSIS
TABLE N - STANDARDIZED INCIDENCE RATIOS (SIRS)' FOR PRIMARY LIVER CANCER AMONG PATIENTS WITH ALCOHOLISM AND LIVER CIRRHOSIS BY SEX AND AGE AT FIRST HOSPITAL ADMISSION FOR ALCOHOLISM OR LIVER CIRRHOSIS (FIRST YEAR OF FOLLOW-UP EXCLUDED) Age at first hospital admission for alcoholism or liver cirrhosis
Men N
~
~
~ Observed , f
Alcoholism cohort 1 3872 < 50 1785 2 50-59 5 1298 60-69 65 1 2 70 + Liver cirrhosis cohort < 50 384 3 50-59 445 9 60-69 528 18 604 7 70 + Alcoholism and liver cirrhosis cohort 285 1 < 50 221 1 50-59 60-69 171 7 57 1 70 +
Expected
Women SIR
95% CI
N:zktPfObserved
Expected
SIR
95% CI
0.5 1.2 1.4 0.8
2.0 1.7 3.5 2.4
0.0-11.1 0.2-6.2 1.1-8.1 0.3-8.8
467 219 133 92
2 0 1 0
0.05 0.1 0.1 0.1
42.8 0.0 11.8 0.0
4.8-154.5 0.0-56.2 0.2-65.6 0.040.4
0.04 0.2 0.4 0.4
82.8 50.9 48.1 17.4
16.6-241.8 23.3-96.8 28.5-76.1 7.0-35.8
278 26 1 402 687
5
4 10 3
0.03 0.1 0.2 0.4
181.7 50.4 44.4 8.4
58.6-424.0 13.6-128.9 21.2-81.6 1.7-24.6
0.0 0.1 0.1 0.03
25.0 10.7 56.0 30.3
0.3-139.1 0.1-59.5 22.4-115.5 0.4-168.6
51 29 13 9
0 1 0 0
0.0 0.01 0.01 0.01
0.0 173.7 0.0 0.0
0.0-762.1 2.3-966.3 0.0-392.9 0.0-380.3
'See footnotes. Table I. TABLE V - STANDARDIZED INCIDENCE RATIOS (SIRS)' FOR PRIMARY LIVER CANCER AMONG PATIENTS WITH LIVER CIRRHOSIS BY ALCOHOLISM STATUS. ACCORDING TO THE 5-DIGIT ICD CODE Men
Women
Observed ExDected SIR
Liver cirrhosis without alcoholism2 Liver cirrhosis with alcoholism3 Liver cirrhosis with unknown alcoholism4
95% CI
Observed Exoected SIR
95% CI
9
0.2
42.1 19.2-80.0
14
0.4
36.9 20.2-62.0
18
0.4
42.0 24.9-66.4
3
0.1
48.8 9.8-142.5
10
0.3
28.7 13.8-52.8
5
0.2
20.1 6.546.9
'Excluding cases of primary liver cancer from the first year of follow-up.-*ICD-7:581.00 or ICD-8571.90 or 571.98; 391 men and 691 ~omen.-~ICD-7:581.01,581.10, 581.20, 581.99 or ICD-8:571.00,571.02,571.09,571.10,571.11,571.19,571.40,571.41; 937 men and 269 ~omen.-~ICD8571.99; 633 men and 668 women. clinical, or it might have been diagnosed previously in another health care region (if the patient has moved into the Uppsala Health Care Region) before the start of the register in 1965 or on an out-patient basis. Under such circumstances, hepatic cirrhosis might not appear as a discharge diagnosis if a patient has been hospitalized, eg., due to an accident. Misclassification of the exposure (alcoholism or cirrhosis) entails underestimation of the true difference in risk between the diagnostic groups. Given the relative magnitude of the respective relative risks, it is evident that the existence of alcoholism per se could have influenced the risk estimates for cirrhosis only marginally. This notion is further supported by the similar excess risk in patients with both diagnoses and in various diagnostic sub-groups of liver cirrhosis (Table V). In contrast, liver cirrhosis might introduce major bias when the association between alcoholism and liver cancer is studied. Indeed, if the relative risk for alcoholism alone is 3 and for liver cirrhosis 36, then 6 percent of undetectable cirrhosis would completely explain the alcoholism association. The impact of confounding by other factors is unknown. In studies with more complete assessment of exposure, confounding seemed to be a minor problem in analyses of alcohol and liver cancer (Tomatis, 1990; La Vecchia et al., 1990), although tobacco smoking, which is more common among alcoholics (Kozlowski and Ferrence, 1990; Cederlof et al., 1975), has been found to be a moderate risk factor for liver cancer (Austin, 1991; Trichopoulos et al., 1987; Hsing et al., 1990). In the present study, the relative risk for lung cancer (excluding the first year and combining sexes) was 2.1 and 1.1among patients with alcoholism and liver cirrhosis, respectively (data not
shown). Hence, tobacco smoking cannot explain the strong associations observed in our study. For liver cirrhosis, the prime candidate as a confounder or an effect modifier of the association between cirrhosis and liver cancer would be chronic active hepatitis-B (Tomatis, 1990; Trichopoulos et aL, 1987) and perhaps hepatitis-C (Kaklamani et al., 1991) infection. The prevalence of the former condition is, however, low in the Swedish population (Olsson et al., 1988), whereas that of the latter is largely unknown. The prevalence of both might be higher in patients with alcoholism and notably liver cirrhosis, on account of these patients' special living conditions, lower socio-economic status and more frequent contact with health care. It is unlikely, though, that the impact of such factors could explain more than a small part of the difference in excess relative risk for primary liver cancer between alcoholism patients with and without liver cirrhosis. In fact, epidemiologic studies in Western populations have in general shown more modest, 10- to 20-fold increases of the risk of primary liver cancer following chronic hepatitis-B infection (Trichopoulos et al., 1987; La Vecchia et al., 1990; Kaklamani et al., 1991; Yu et al., 1983) than we found in patients with liver cirrhosis. This reasoning leads us to the tentative conclusion that alcohol and even alcoholism leads to liver cancer mainly (perhaps exclusively) through development of liver cirrhosis as an intermediate step. With regard to liver cirrhosis and liver cancer, methodologic concerns may be less serious and causal interpretation more straightforward. Since ascertainment of both hepatic and biliary-tract cancer might be more complete in patients with cirrhosis, thereby exaggerating relative risk estimates,it was reassur-
902
ADAMI ET AL.
ing to find little or no evidence of excess risk in relation to biliary cancer. In Sweden, as in Denmark (Skinhoj and Prytz, 1981), alcoholism and unknown (“idiopathic”) causes likely account for most cases of livcr cirrhosis, whereas the proportion due to chronic viral hepatitis as suggested above is low. Previous follow-up studies of primary liver canccr in paticnts with alcoholism were small, with few observed cases and generally reported risks about 50% higher than in nonalcoholics (Tomatis, 1990; Austin, 1991; Hardell et al., 1984; Trichopoulos et al., 1987; Tuyns, 1982; Falk, 1982; Babor et al., 1987). Conceivably, our higher estimates may b e due to more severe and long-lasting abuse (and consequently a higher prcvalcnce of liver cirrhosis) in those given a discharge diagnosis of alcoholism. This may pertain especially to women, for whom social barriers entail a more restrictive use of alcoholism as a diagnosis. However, even in men the diagnosis is restricted to those with obvious alcoholism and not uscd for patients who have accidentally abused. Tndeed, underdiagnosis of alcoholism in the dischargc rcgister is likely to be substantial. Risk estimates for livcr cancer among patients with liver cirrhosis provided by case-control studies are generally lower
(Trichopoulos et al., 1987; La Vecchia et al., 1990), cvcn whcn no latency period was applied to avoid sclection bias (La Vccchia et al., 1990). Prospective studies in paticnts with cirrhosis have bccn small in siLe (Zaman et al., 1985; Hadengue et al., 1990; Kobayachi et al., 1990; Oka et al., 1990; Borzio et al., 1991) and quantitative risk estimates were not always rcportcd (Kobayachi et al., 1990; Oka et al., 1990). Our data did not support previous claims that mcn (Zaman et ul., 1985; Kobayashi et al., 1990) or age over 60 (Borzio et UL, 1991) are effect modifiers for liver cancer in paticnts with cirrhosis. It has been reported that 10 to 30% of patients with cirrhosis develop liver cancer (Rothman, 1980); that the cumulative incidence after 6 years is 39% (Oka et al., 1990) and that 213 (Okudaet al., 1987) of patients with liver cirrhosis in Japan dic from liver cancer. These extreme rates clearly did not pertain to the Swedish population (Fig. 1 ) . ACKNOWLEDGEMENTS This study was supported by grants from the Swedish Cancer Socicty. W e thank Mr. E. Berger for excellent assistance with the data analyses.
REFEKENCLS ALSTIN.H., The role of tobacco use and alcohol consumption in the etiology of hepatocellular carcinoma. In: E. Tabor. A.M. Di Bisceglie and R.H. Purcell (eds.), Etiology, pathology, and treatment ofhepatocellular carcinoma in North America. Houston: Adv. appl. Biotechnol.,13, pp. 57-75 (1991). BABOR, T.. STEVENS, R. and MARLATT, A,, Verbal report methods in clinical research on alcoholism: response bias and its minimization.J. Stud. Alcohol, 48,410-424 (1987). BORZIO,M., BRUNO, S., RONCALLI, M., MELS,G.C., RAMELLA, G., BORLIO,F., LEANDRO, G. and PODDA,M., Liver cell dysplasia and risk of hepatocellular carcinoma in cirrhosis: a preliminary report. Brit. med. J., 302,1312 (1991). CEDERI.~F, R., FRIBERG, L., HRUBEC,Z. and LOKICH,U., The relationship of vmoking and some social co-variables to mortality and cancer morbidity. The Department of Environmental Hygiene, the Karolinska Institute, Stockholm (1975). FALK, H.. Liver. In: D. Schottenfeld and J.F. Fraumeni (eds.), Cancer epidemiology and prevention, pp. 668-682, Saunders, Philadelphia (1982). HADENG~JE, A., N’DRI, N. and BENHAMON, J.P., Relative risk of hepatocellular carcinoma in HBsAg positive vs. alcoholic cirrhosis. A cross-sectional study. Liver, 10, 147-151 (1990). HARDELL, I ,., BENGTSSON,N.O., JONSSON, U., ERIKSSOK. S. and LARSSON, L.G., Aetiological aspects on primary liver cancer with special regard to alcohol, organic solvents and acute intermittent porphyria- an epidemiological investigation. Brit. J. Cancer, 50, 389397 (1984). HSING,A.W.. MCLAUGHLIN, J.K., HRUBEC,Z., BLOT,W.J. and FRAUMENI, J.F., Cigarette smoking and liver cancer among US veterans. Cancer Causes Control, 1,217-221 (1990). INTERNATIONALA ~ ~ L NFOR C YRESEARCH ON CANCER,Alcohol drinking. IARC Monographs on the evaluation of carcinogenic risks, 44, TARC, Lyon (1988). KAKLAMANI. E.. TRICHOPOULOS, D., Tzon-ou, A., ZAVITSANOS, X.. KOUMANTAKI, Y., HKIZAKIS. A., HSIEH,C.C. and HXI‘ZIYANILIS, S., Hepatitis B and C viruses and their interaction in the origin of hepatocellular carcinoma.J. Anter. med. Ass., 265, 1974-1976 (1991). KOBAYASHI, K., UNOURA,M.? TAYAKA, N. and HATTORI, N., A comparison bctwccn hepatocellular-carcinoma-developing and noncarcinoma-developing patients with cirrhosis over a long follow-up 37,445-448 (1990). period. Hepatogastroenter[~lo~, KOZLOWSKI, L.T. and F~KKE~YICE, R.G., Statistical control in research on alcohol and tobacco: an example from research on alcohol and mortality. Brit. J. Addiction, 85,271-278 (1990).
LA VECCHTA, C., NEGRI,E., D’AvA~Lo,B., BOYLE, P. and FRANCESCHI,S., Medical history and primary liver cancer. Cancer Rex, 50, 6274-6277 (1990). NAESSEN, T., PARKER, R., PERSSON, 1.; ZACK,M. and ADAMI, H.O., Time trends in incidence rates of first hip fracture in the Uppsala Health Care Region, Sweden 1965-1983. Amer. J. Epidemiol., 130, 289-299 (1989). OM, €I,, KLRIOKA,N., KIM,K., K?L”O,T., KUROKJ.T., MIZOGUCHI, Y. and KOBAYASHI; K.; Prospective study of early detection of hepatocellular carcinoma in patients with cirrhosis. Hepatology, 12, 680-687 (1990). OKUDA,K.: FUIIMOTO, I., HANAI,LA. and URANO,Y.. Changing incidence of hepatocellular carcinoma in Japan. Cancer Rex, 47, 49674972 (1987). OLSSON. R., LINDBERG, J., WEILANU, O., NILSSON,L. and members of the Swedish Internal Medicine Club and Clinics for Infectious Diseases, Chronic active hepatitis in Sweden. Scund. J. Gastroenterol., 23, 463470 (1988). ROTHMAN, K.J., The proportion of cancer attributable to alcohol consumption.Prev. Med., 9,174-179 (1980). SKINHBJ, P. and PRYTZ,H., Changing mortality from cirrhosis in Denmark 1965-1978. Scund. J. Gastroenterol., 16,833-837 (1981). SWEDISH CANCER REGISTRY, Cancer incidence in Sweden I965-1984. National Board of Health and Welfare (Annual publications) Stockholm, Sweden (1969-1987). TOMATIS, L., Cancer cuuses, occurrence and control. IARC Scientific Publication 100, IARC, Lyon (1990). TRICHOPOULOS, D., DAY,N.E., KCIKLAMANI, E., TZONOU, A.! MuGoz, N., ZAVIISANOS, X., KOUMAN IAKI, Y. and TRICHOPOULOS, A,, Hepatitis B virus, tobacco smoking and ethanol consumption in the etiology of hepatocellular carcinoma. Int. J. Cancer, 39,45-49 (1987). TLYNS,A.J.. Alcohol. In: D. Schottenfeld, D. and J.F. Fraumcni (eds.), Cancer epidemiology and prevention, pp. 293-303, Saunders, Philadelphia (1982). WHELAN, S.L., PARKIN, D.M. and MASUYF,R, E. (eds.), Patterns of cancer in$ve continents. IARC ScientificPublication 102, IARC, Lyon (1990). Yu, M.C., MACK,T., IIANISCH? R., PETERS,R.L., HENDERSON, B.E. and PIKE,M.C., Hepatitis, alcohol consumption, cigarette smoking, and hepatocellular carcinoma in Los Angeles. Cancer Rex, 43, 60776079 (1983). ZAMAN, S.N., MELIA,W .M.,JOHNSON. R.D., PORTMAN, R.C., JOHKSON, P.3. and WILLIAMS, R., Risk factors in development of hepatocellular carcinoma in cirrhosis. Prospective study of 613 patients. Lancet, I, 1357-1360 (1985).
